Benchmark Heaven
Price & provider filters · adjusted costs
Global
Applies to price views & model offers; benchmark evidence stays unfiltered
← All models

DeepSeek V4 Pro (Non-reasoning)

open weights★ featured
DeepSeek · released 2026-04-24 · 18 recorded token offers

Top 5 cheapest providers (Adjusted $/task)

Within the active global provider, residency and confidentiality filters.

#ProviderPlatformRaw input $/1MRaw output $/1MAdjusted $/task
1DigitalOceanOpenRouter$0.870$1.74
2GMICloudOpenRouter$0.957$1.91
3FireworksOpenRouter$1.20$1.20
4DeepInfraOpenRouter$1.30$2.60
5NovitaOpenRouter$1.60$3.20

Benchmarks

Composite53.8
Mean-imputed base42.5
Dominance adjustment+11.3
Composite evidence1/5
AA Coding · unversioned snapshot 2026-09-10
AA Coding Agent · v1.4 · 2026-09-09
AA Intelligence · unversioned snapshot 2026-09-1020.8
DesignArena Frontend · unversioned snapshot 2026-09-10
DesignArena Full-Stack · unversioned snapshot 2026-09-10
Output speed (t/s)60

MODEL EVIDENCE

Benchmark sheet

8 / 73 registered benchmark versions covered · 9 observations including existing index snapshots.

Compare this model ↗
Composite retains Coding Agent v1.4 · source 2026-09-09

The five Composite inputs remain unchanged. Its Coding Agent input is the median across complete harness results in the retained v1.4 snapshot from 2026-09-09. Artificial Analysis now publishes v1.5, with different components. Explore current v1.5 separately. Dated snapshot labels on other indices identify unversioned source captures, not a verified semantic version.

Profile signals

No threshold-crossing signal · 9 eligible benchmark families

How flags are calculated

Heuristic screen, not statistical significance: benchmark families are correlated and source uncertainty is unknown. Peer evidence needs ≥ 20 independently measured matched configurations from ≥ 10 distinct model families. A flag needs a directed population z-score of magnitude ≥ 1.5 and a gap of ≥ 1.5 from the leave-one-benchmark-family-out mean z in the same direction, over ≥ 5 other benchmark families.

Protocol-compatible measured/vendor divergences

No verified protocol-compatible vendor/measured pair is available for this model; agreement cannot be assessed.

Agentic

Agentic: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
Terminal-Bench Hard (AA)

Version 74221fb

Published board

Tests a pinned 44-task hard subset of terminal-based work using Terminus 2.

0.36364 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: Terminal-Bench Hard (AA) · 74221fb · Published board
Exact value: 0.363636363636364 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:terminalbenchHard

Instruction-following

Instruction-following: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
IFBench (AA single-turn)

Version snapshot-2026-09-10

Published board

Tests precise single-turn instructions with deterministic rule checks.

0.45782 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: IFBench (AA single-turn) · snapshot-2026-09-10 · Published board
Exact value: 0.457823129251701 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:ifbench

Knowledge

Knowledge: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
Humanity's Last Exam (AA text-only)

Version snapshot-2026-09-10

Published board

Tests expert-level knowledge on AA's text-only Humanity's Last Exam subset.

0.08248 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: Humanity's Last Exam (AA text-only) · snapshot-2026-09-10 · Published board
Exact value: 0.082483781278962 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:hle
AA-Omniscience Index

Version snapshot-2026-09-10

Published board

Tests factual reliability while rewarding correct answers and penalizing hallucinations.

-29.86667 pointsmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: AA-Omniscience Index · snapshot-2026-09-10 · Published board
Exact value: -29.8666666666667 points
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:omniscience

Long-context

Long-context: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
AA-LCR v1.1

Version 1.1

Published board

Tests reasoning across multiple long documents with corrected answer keys and grading.

0.53 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: AA-LCR v1.1 · 1.1 · Published board
Exact value: 0.53 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:lcr

Reasoning

Reasoning: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
AA Intelligence Index

Version snapshot-2026-09-10 (unversioned)

Published board

Published AA index from the retained API snapshot. No verified semantic version was supplied.

20.8 pointsmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: AA Intelligence Index · snapshot-2026-09-10 (unversioned) · Published board
Exact value: 20.8 points
Observed: 2026-09-10 · publication date: not recorded
Observation id: legacy:aa_intelligence_index:deepseek-v4-pro::non-reasoning
Protocol: inspect · file data/raw/artificialanalysis.json

Science

Science: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
CritPt (AA)

Version snapshot-2026-09-10

Published board

Tests research-level physics reasoning with Python, symbolic and numerical answers.

0.00857 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: CritPt (AA) · snapshot-2026-09-10 · Published board
Exact value: 0.00857142857142857 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:critpt
GPQA Diamond (AA)

Version snapshot-2026-09-10

Published board

Tests graduate-level biology, physics and chemistry knowledge on the Diamond subset.

0.71717 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: GPQA Diamond (AA) · snapshot-2026-09-10 · Published board
Exact value: 0.717171717171717 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:gpqa

Tool-use

Tool-use: every observation for this exact model configuration
Benchmark / versionEvaluation groupScore / provenance
τ²-Bench Telecom (AA)

Version snapshot-2026-09-10

Published board

Tests dual-control telecom agents that coordinate tool use with a simulated user.

0.91228 fractionmeasuredobserved 2026-09-10artificialanalysis.ai
Evidence
Axis: τ²-Bench Telecom (AA) · snapshot-2026-09-10 · Published board
Exact value: 0.912280701754386 fraction
Observed: 2026-09-10T21:47:16.627Z · publication date: not recorded
Observation id: aa:6e2a2572-ed8d-4616-8d7d-68d125fc8ee7:tau2
Missing coverage · 68 benchmark versions

No result does not mean a zero, or that the model was never tested. Collection failures and disputed versions retain their distinct status.

Variants / reasoning settings

VariantCoding · snapshot 2026-09-10Intelligence · snapshot 2026-09-10
DeepSeek V4 Pro (Reasoning, Max Effort)59.430.9
DeepSeek V4 Pro (Reasoning, High Effort)58.730.1
DeepSeek V4 Pro (Non-reasoning)20.8

Token offers by platform — Adjusted $/task

Adjusted costs are modeled USD per task: AA output tokens × OpenRouter usage I/O (Chutes global fallback). These are general usage and benchmark proxies for coding-agent work. Missing AA data assumes 1,000 output tokens/task; unknown cache hit assumes 0%; unmeasured additional cache writes assume 0 tokens. Click any underlined price for exact inputs, dates and assumptions. Raw list prices use the selected fixed input/output blend, in USD per million tokens.

14 offers within the active global filters; “—” means the catalog is active but no public token price is available.

OpenRouter (11)

DigitalOceanglobaldigitalocean$0.870 raw in $/1M$1.74 raw out $/1M
GMICloudglobalgmicloud/fp8$0.957 raw in $/1M$1.91 raw out $/1M
DeepInfraglobaldeepinfra/fp8$1.30 raw in $/1M$2.60 raw out $/1M
Novitaglobalnovita/fp8$1.60 raw in $/1M$3.20 raw out $/1M
Veniceglobalvenice$1.65 raw in $/1M$3.30 raw out $/1M
AtlasCloudglobalatlas-cloud/fp4$1.68 raw in $/1M$3.38 raw out $/1M
BaseTenglobalbaseten/fp4$1.74 raw in $/1M$3.48 raw out $/1M
Parasailglobalparasail/fp8$1.74 raw in $/1M$3.48 raw out $/1M
NextBitglobalnextbit/fp8EU$1.80 raw in $/1M$3.55 raw out $/1M
Azureglobalazure/us$1.91 raw in $/1M$3.83 raw out $/1M
Fireworksglobalfireworks$1.20 raw in $/1M$1.20 raw out $/1M

Azure AI Foundry (2)

Azure AI Foundryglobalazure directEU equivalent$1.74 raw in $/1M$3.48 raw out $/1M
Azure AI Foundryusfireworks$1.93 raw in $/1M$3.83 raw out $/1M

Nebius (1)

Nebiusuk$1.75 raw in $/1M$3.50 raw out $/1M
Benchmark Heaven — Model benchmarks & costs